sequence-load（顺序加载）

Sequence Load Skill

<quick_start> Trigger: Monday 7:15 AM ET (runs after prospect-refresh at 6:30 AM) Manual Trigger: "Load prospects into sequences" or "Enroll new leads" Dependencies: Requires Apollo credits, prospect-refresh output, HubSpot portal 21530819 Output: Enrollment confirmation per prospect, sequence assignment, contact IDs for HubSpot creation </quick_start>

<success_criteria>

• Read prospect-refresh output (30 net-new prospects from Gmail drafts + HTML report)
• Query HubSpot to verify prospect not already enrolled in any sequence
• Validate phone number exists (from prospect-enrich or Apollo)
• Apply Golden Rules (skip customers, channel, device owners, NEVER ATL)
• Find target Apollo sequence(s) by ICP vertical
• Get email account ID (tim@epiphan.com or primary)
• Batch add prospects to sequence (max 100 per request)
• Confirm enrollments (check Apollo contact.sequence_id)
• Create HubSpot contacts if not exist (pull contact IDs for HubSpot sync)
• Report: total enrolled, per-sequence breakdown, any failures </success_criteria>

Stage 1: Read Prospect Refresh Output

Input Source: prospect-refresh HTML report + Gmail drafts

Extract from HTML Report:

• Prospect name, email, title, company, vertical, ATL/BTL tier
• Apollo person_id, organization_id, phone
• Company revenue, headcount, funding

Validation:

• Email format is valid
• Phone exists (from enrichment)
• Vertical is one of: Higher Ed, Courts, Gov, Corporate AV, Healthcare, HoW, K-12

Output: Structured list of 30 prospects (or fewer if filtering applied)

Stage 2: Deduplicate Against HubSpot + Apollo

Step A: Check HubSpot for existing contact

MCP Tool: search_crm_objects (HubSpot)

objectType: "contacts"
filterGroups: [{
  filters: [
    { propertyName: "email", operator: "EQ", value: prospect.email }
  ]
}]
properties: ["email", "phone", "hs_lead_status"]

Decision Logic:

• If contact exists + NOT in "Sequenced" or "Active Sequence" status → REUSE contact (get contact ID)
• If contact exists + in sequence → SKIP (avoid re-enrollment)
• If contact does not exist → PROCEED to Stage 3 (create new)

Step B: Check Apollo for duplicate enrollments

MCP Tool: apollo_contacts_search

q_keywords: prospect.email
per_page: 10

Decision Logic:

• If Apollo contact exists + already in a sequence → SKIP
• If Apollo contact exists + no sequence → PROCEED (re-enroll)
• If Apollo contact does not exist → PROCEED to Stage 3 (create new)

Dedup Output: List of unique prospects (typically 60-70% of input after dedup)

Stage 3: Validate Golden Rules + Phone Exists

Golden Rules (Hard Exclusions):

Skip if:

• Email domain matches known customer domain (cross-check crm_customers)
• Contact tagged "Channel Partner" in HubSpot
• Contact tagged "Device Owner" in HubSpot
• Company in "Product Page Engagers" tag
• Title in NEVER ATL list: Warehouse Manager, Network Manager, Systems Admin, AV Technician, Graphic Design Instructor, Program Administrator, Web Designer, Classroom Support, Lab Coordinator, Maintenance, Building Engineer, Multimedia Services Manager, Video Production Specialist, Streaming Crew

Phone Validation:

• Phone must be non-null (from Apollo enrichment or prospect-enrich)
• Format: +1-XXX-XXX-XXXX (US) or +1-XXX-XXX-XXXX (Canada)
• If phone missing → LOG as "Phone Missing" but DO NOT ENROLL (requires phone for Apollo compliance)

Output: Verified prospect list (typically 80-90% after phone validation)

Stage 4: Map Prospects to Target Sequences

MCP Tool: apollo_emailer_campaigns_search

Search for sequences by vertical + template:

Query Logic:

per_page: 100
# Search returns: id, name, status, num_contacts, num_enrolled

Sequence Selection Rule:

• Map prospect.vertical to sequence_name_pattern
• Choose sequence with LOWEST current enrollment (load balancing)
• Fallback: default "BDR_Outbound_General" if vertical-specific not found

Output: Prospect + sequence_id mapping

Stage 5: Create Apollo Contacts (If New)

MCP Tool: apollo_contacts_search

For new prospects (not found in Stage 2):

per_page: 10
q_keywords: prospect.email

If not found, create contact via:

MCP Tool: apollo_contacts_create

email: prospect.email
first_name: prospect.first_name
last_name: prospect.last_name
title: prospect.title
organization_name: prospect.company
phone: prospect.phone (if available)
label_names: [prospect.vertical, "BDR_Prospect", "2026_Q1"]

Output: New Apollo contact ID (apollo_contact_id)

Stage 6: Get Email Account ID

MCP Tool: apollo_email_accounts_index

Query all linked email accounts:

# Returns: [{ id, email, status, daily_send_limit, ... }]

Selection Logic:

• Look for email containing "epiphan" or "tkipper"
• Use primary email account if multiple
• If none found → ERROR (cannot proceed without email account)

Output: email_account_id (e.g., "6633baaece5fbd01c791d7ca")

Stage 7: Batch Add to Sequences

MCP Tool: apollo_emailer_campaigns_add_contact_ids

For each sequence (group prospects by sequence_id):

id: sequence_id (from Stage 4 mapping)
emailer_campaign_id: sequence_id (same as id)
send_email_from_email_account_id: email_account_id (from Stage 6)
contact_ids: [contact_id_1, contact_id_2, ..., contact_id_N] (max 100 per batch)
sequence_active_in_other_campaigns: true (allow multi-sequence enrollment)
sequence_unverified_email: false (require verified emails)
status: "active" (start sequence immediately)

Batch Logic:

• Group prospects by target sequence
• Send 100 contacts max per API call
• Retry if rate limited

Expected Success Rate: 95%+ (most failures due to email validation)

Stage 8: Confirm Enrollments

MCP Tool: apollo_contacts_search

For each enrolled contact, verify enrollment:

q_keywords: contact.email
per_page: 1

Check response:

• contact.sequences (should contain sequence_id from Stage 7)
• contact.last_sequence_enrollment_date (should be "today")
• contact.status (should be "active")

Log Enrollment:

✓ jane@acme.com → BDR_HigherEd_Outbound (sequence_id: abc123)
✗ bob@example.com → Phone validation failed
✗ carol@company.com → Already enrolled in BDR_Corporate_AV

Stage 9: Create HubSpot Contacts (Sync)

For newly created Apollo contacts, create HubSpot counterparts:

MCP Tool: manage_crm_objects (HubSpot createRequest)

objectType: "contacts"
properties: {
  firstname: prospect.first_name,
  lastname: prospect.last_name,
  email: prospect.email,
  phone: prospect.phone,
  jobtitle: prospect.title,
  company: prospect.company_name,
  lifecyclestage: "subscriber",
  hs_lead_status: "open",
  custom_apollo_contact_id: apollo_contact_id,
  custom_apollo_sequence: sequence_name,
  custom_atl_btl_tier: prospect.atl_btl_tier,
  custom_prospect_vertical: prospect.vertical,
  custom_prospect_icp_score: prospect.icp_score
}
associations: [
  {
    targetObjectId: company_id, # Find or create company in HubSpot
    targetObjectType: "companies"
  }
]

Company Association Logic:

• Search HubSpot for company by domain
• If found: use company_id from existing record
• If not found: create company via manage_crm_objects

Output: Created contact IDs (to link Gmail drafts)

Stage 10: Output Enrollment Report

Format: Markdown table + summary stats

Enrollment Table:

Summary Stats:

• Total prospects processed: X
• Successfully enrolled: Y (Y% of X)
• Phone validation failures: Z
• Duplicate/already enrolled: W
• HubSpot synced: Y (new contacts)
• Per-sequence breakdown:

- BDR_HigherEd_1: 5 enrolled - BDR_Courts_1: 4 enrolled - BDR_CorporateAV_1: 5 enrolled - etc.

Next Steps: Prospects now enrolled in Apollo sequences. Gmail drafts (from prospect-refresh) ready for manual review + send-from-draft workflow.

Emit Outcome Sidecar

As the final step, write to ~/.claude/skill-analytics/last-outcome-sequence-load.json:

{"ts":"[UTC ISO8601]","skill":"sequence-load","version":"1.0.0","variant":"default",
 "status":"[success|partial|error]","runtime_ms":[estimated ms from start],
 "metrics":{"contacts_loaded":[n],"sequences_updated":[n],"duplicates_skipped":[n],"phone_validation_failures":[n]},
 "error":null,"session_id":"[YYYY-MM-DD]"}

Use status "partial" if some stages failed but results were produced. Use "error" only if no output was generated.

Skill Metadata

Version: 1.0 Last Updated: 2026-03-19 Author: Tim Kipper Status: Production Integration: Apollo + HubSpot (portal 21530819) Tier: P1 (Core BDR Automation) Triggers: Scheduled (Monday 7:15 AM) + Manual ("Load sequences") Dependencies: prospect-refresh (6:30 AM) → sequence-load (7:15 AM) → morning-brief (7:30 AM)